{"generatedAt":"2026-10-08T02:25:04.299Z","methodology":"Two-year daily log returns. Discovery is the first 65% of the calendar. Every directed asset-pair × 1–5 trading-day lag is a separate hypothesis. Raw Student-t p-values are corrected across the full search space with Benjamini-Hochberg FDR at 5%. Surviving hypotheses must then retain sign and effect out-of-sample over the final 35%, split into three forward folds.","caveat":"Lead/lag remains observational. FDR controls expected false-discovery proportion under its assumptions; it does not prove causality, persistence, tradability or net profitability after costs.","universeSize":31,"validation":{"dataWindow":"2y daily","discoveryFraction":0.65,"discoveryCutoff":"2026-01-25","oosStart":"2026-01-26","hypothesesTested":4650,"fdrMethod":"Benjamini-Hochberg","fdrAlpha":0.05,"bonferroniAlpha":0.000010752688172043012,"significantAfterFdr":0,"walkForwardFolds":3,"minimumDiscoveryObservations":90,"minimumOosObservations":45,"stabilityDefinition":"0.6 × fraction of OOS folds matching the discovery sign + 0.4 × min(1, |OOS r| / |discovery r|). Required stability ≥ 0.65.","publishedRelationships":0},"relationships":[]}